The post announces the 2026-07-28 Model Context Protocol specification, with the headline change that HTTP transport is now stateless instead of relying on server-side sessions. In plain terms, MCP is a way for AI clients like coding assistants and chat agents to call external tools and services. The old transport model made ordinary web deployment harder than it needed to be. It forced servers and intermediaries to keep per-client state, which clashed with standard HTTP patterns and made serverless hosting, retries, scaling, and failure handling more painful.
That landed as the clear win here. People running MCP infrastructure said session persistence caused a huge share of real bugs and operational burden. Several developers said they had already moved their own setups toward plain HTTP and saw reliability improve. The practical read from the comments is that MCP is shedding a self-inflicted protocol mistake and moving closer to the way web systems already work.
The catch is migration. This is not being treated like a cosmetic change. One operator with a large index of open source MCP servers said the new protocol is
wire-incompatible both ways, so many clients and servers need refactoring, not just
SDK upgrades. That same comment put some scale on the ecosystem, claiming roughly 18,700 of 62,700 indexed open source servers had been updated in the last 30 days. That suggests there is enough active surface area for the transition to be noisy. It also creates a near-term opening for MCP gateways that translate between old and new implementations.
The other strong signal is what still is not solved. File transfer came up immediately. Developers described clients stuffing
base64-encoded binary data into prompts, truncating image payloads, or handling file inputs inconsistently. Maintainers acknowledged file uploads were deferred from this release and remain unfinished. So while stateless transport fixes the deployment model, MCP still has obvious product gaps for any workflow that needs to move images or other binary files cleanly.
The mood was mostly relieved and positive. Even some skeptics framed this as MCP finally converging on familiar HTTP behavior. The remaining friction is less about whether stateless transport is right and more about how painful the compatibility window will be, how quickly major clients adopt the new spec, and whether the protocol can close basic holes like file transfer before teams decide plain HTTP plus documentation is simpler.
blog.modelcontextprotocol.io